The question comes up often: do you really use Claude differently depending on your subscription? On Max $200, should you use Opus at maximum effort all the time, or is there a smarter way to work?

Short answer: yes, the workflow changes. Not just because of the model — because of budget psychology. When you know you can burn through your daily limit in 30 minutes, you code differently. And when the limit disappears, other things become important instead.

What the three subscriptions actually give you

Before talking workflow, the raw facts. Anthropic talks about "5x usage" for Pro, "20x" for Max $100, "60x" for Max $200. In practice with Claude Code:

Pro ($20)

Max ($100)

Max ($200)

Default model

Sonnet

Sonnet or Opus

Opus

Long sessions

Avoid

OK

No constraint

/clear between tasks

Required

Recommended

For speed

Optimised CLAUDE.md

Critical

Useful

Useful (speed)

Parallel agents

Conserve

OK

Generously

Full audits / reviews

Rare

Regular

At will

Pro ($20) — working with constraints, not against them

On Pro, Opus is available but rationed. You get it a few times per day, and an intense session can exhaust it. Sonnet is the everyday model.

The natural reflex is to "save Opus for real questions." That's partially right, but badly applied it becomes paralysing — you spend time deciding whether the question deserves Opus instead of actually working. The useful rule is simpler: Sonnet by default, Opus only when the question involves an irreversible decision or a bug you can't diagnose after two attempts.

What genuinely changes the value/token ratio on Pro:

/clear between tasks, not end of day. A session that accumulates context across 3 different topics costs 3x more than 3 clean sessions. The discipline here is real — you need to stop, clear, restart. It goes against the feeling of "momentum" but the difference on the day's limit is significant.

Broad questions are expensive for little return. "How can I improve this file?" triggers a complete analysis. "The buildPayload() function on line 47 — does it handle the null case?" costs 10x less and gives a more useful answer. Precision isn't just courtesy — it's economics.

Short CLAUDE.md, complete .claudeignore. Every line of CLAUDE.md is loaded every session. A 300-line file versus a 25-line file, multiplied across all sessions in a day, represents a non-negligible fraction of your budget. The token optimisation covered in this article isn't optional on Pro.

Max ($100) — the middle ground that enables real sessions

The jump from Pro to Max $100 is bigger than it looks. It's not just "more tokens" — it's the disappearance of defensive planning. You stop asking yourself whether you can afford this question.

Opus is accessible for serious work sessions. Sonnet stays relevant for mechanical tasks: generating a commit message, making a quick fix, answering a documentation question. The switch doesn't have to be conscious every time — Sonnet by default, Opus when you know you're tackling something complex.

What Max $100 unlocks concretely:

Full reviews and audits. On Pro, asking for a complete review of a 500-line module is a luxury. On Max $100, it's a regular practice. The quality of code on an active project reflects this directly.

Long refactoring sessions. A refactor touching 5 files with cross-dependencies requires Claude to hold the full context. On Pro, you fragment and lose coherence. On Max $100, you can run the session in full.

Parallel agents without guilt. Launching two agents in parallel to explore two approaches is reasonable on Max $100. On Pro, it doubles your usage in one shot.

Max ($200) — when the constraint disappears, what remains

Permanent Opus, no counting. The question becomes: do you really need to run everything at maximum effort?

No. Not for budget reasons, but for speed and clarity reasons. A context loaded with 200 files and a 2-hour conversation gives slower and sometimes less precise answers than a clean session focused on one problem. The discipline of /clear stays useful — not to save tokens, but because a short context remains better than an overloaded one.

What Max $200 changes in practice:

Systematic audits. Before starting a feature, have Claude read the entire relevant module to understand existing patterns. On lower subscriptions, that's an investment you ration. On Max $200, it's the natural starting point.

Generous parallel agents. Launching 3 or 4 agents on independent tasks simultaneously — exploring approaches, generating tests, writing documentation — without watching the counter. The real time savings on complex projects is substantial.

Extended research mode. Two-hour sessions on an architecture problem, with multiple back-and-forths, multiple code reads, multiple tested proposals. The kind of work you avoid on Pro because it's too expensive, but that produces the best decisions.

What stays useful regardless of subscription

Two things don't change with budget:

Question precision. "Look at the whole codebase and tell me how to improve the architecture" is a bad question on any subscription. Not because it's expensive, but because it produces a vague answer. Precision is a communication skill, not an economics skill.

Short CLAUDE.md. On Pro, it saves budget. On Max $200, it saves speed — less useless context loaded each session, faster responses, less noise in the instructions. A 25-line targeted CLAUDE.md remains better than a 200-line exhaustive one, regardless of what the subscription allows.

Conclusion

The real determinant of workflow isn't the model — it's the presence or absence of the budget constraint. On Pro, it structures everything: questions, session length, which tasks to delegate. On Max $200, it disappears and other priorities emerge — speed, precision, context quality.

What doesn't change: the basic good practices. Short CLAUDE.md, precise questions, /clear between distinct contexts. These habits aren't economy hacks — they produce better interactions regardless of budget. They apply on Pro out of necessity, on Max by choice. The result is the same.